[Intel-gfx] [Mesa3d-dev] [RFC] reduce number of visuals exposed by Intel drivers

Ian Romanick idr at freedesktop.org
Mon Feb 8 18:52:32 CET 2010


-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Pauli Nieminen wrote:
> On Sat, Feb 6, 2010 at 4:35 PM, Eric Anholt <eric at anholt.net> wrote:
>> On Fri, 05 Feb 2010 13:01:50 -0800, Ian Romanick <idr at freedesktop.org> wrote:
>>> -----BEGIN PGP SIGNED MESSAGE-----
>>> Hash: SHA1
>>>
>>> I'd like to remove a bunch of the visuals and fbconfigs exposed by the
>>> Intel drivers.  There are several categories of visuals that are likely
>>> not useful to anyone.  A couple of our test suites (e.g., glean) like to
>>> run over every possible visual.  As a result, the test suites take an
>>> extraordinary amount of time to run.
>>>
>>> I propose removing:
>>>
>>>   * All 24-bit depth / 0-bit stencil visuals.  These are compatible with
>>> the default state of a 24-bit depth / 8-bit stencil visual and offer no
>>> memory savings.  This will eliminate 24 (of 72) visuals by itself.
>>>
>>>   * All but one of the visuals with accumulation buffer.  Accumulation
>>> is a software path in the Intel drivers (though this could be fixed), so
>>> I don't see any utility in offering multiple, optimized buffer
>>> configuration choices.  This will eliminate an additional 23 visuals.
>>>
>>> This will leave the 25 visuals and 37 fbconfigs that are likely to be
>>> useful.
>> Yes!
> 
> Wouldn't it be better to make piglit default to --quick for glean
> tests? Then add a switch to piglit runner that removes the quick
> option.

piglit isn't the problem.  It's glean and an internal closed-source test
suite that are the problem.  We don't have particular control over
either one.

> If quick option doesn't test enough configs yes glean could select
> some 2 to 4 configs to test.

Remember the golden rule of testing: anything that isn't tested is
broken.  We've had features like that in the past, and every now and
then one sort of visual breaks and goes unnoticed for a long time.

> But I don't oppose cleaning the fbconfig list. Specially when their
> value is questionable.

That was my thinking.  It's also worth noting that I'm the one that
first changed the drivers to expose all these useless visuals anyway.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.10 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/

iEYEARECAAYFAktwT10ACgkQX1gOwKyEAw+xNwCgj3l86QtP628ZjNbM0M0uSJq+
cqYAoIaWUygc5VcFRjQAugS37ofVQaU6
=ylVs
-----END PGP SIGNATURE-----



More information about the Intel-gfx mailing list